Automatic recognition and parameter extraction in GPR images

In this paper, we propose a fast and efficient method for fitting dual-region wave recognition to GPR images under real conditions. Firstly, the GPR image is segmented into hyperpixels according to the idea of "points instead of regions". Then, a novel multi-scale neighborhood computation method is proposed to obtain the region of interest to achieve the binarization of the source image. Subsequently, skeleton extraction was performed in the region of interest by opening and closing and other algorithms. Finally, the region of hyperbolic waves is selected and its corresponding equations are obtained by matching the downward opening line segments based on the known fitting set, and the accuracy of the extraction is judged. In the proposed method, the multiscale neighborhood computation to obtain the region of interest and the final matching algorithm are its two core parts. The preliminary multi-scale neighborhood computation and the final matching algorithm take into account the geometrical characteristics of the two-region wave, which makes the algorithm more robust. The paper confirms the effectiveness of this algorithm through experiments.